OpenAI Confirms ChatGPT Messages May Be Reviewed and Reported to Police

OpenAI has clarified how ChatGPT handles sensitive conversations, especially those involving self-harm or threats to others. In a recent blog post, the company explained the safeguards in place, when human moderators may step in, and the rare scenarios where police could be alerted.

Drawing the Line: Self-Harm vs. Harm to Others

OpenAI says ChatGPT is built to provide empathetic support when people share feelings of distress. For example, if a user expresses suicidal thoughts, the AI responds with compassion and directs them to professional resources such as 988 in the US or Samaritans in the UK. Crucially, these conversations are not automatically escalated to law enforcement, a step OpenAI avoids in order to protect users’ privacy.

The situation changes if a person threatens someone else. In such cases, the conversation is passed into a special review pipeline, where trained moderators examine the details. If they determine that the threat is imminent and real, OpenAI may contact law enforcement. Accounts linked to such cases can also be suspended or banned to prevent further misuse.

Weaknesses in Long Conversations

The company acknowledged that while its safety systems work well in short interactions, they may become less effective in long or repeated conversations. Over time, the AI’s safeguards can degrade, raising the risk of responses that don’t fully align with safety protocols. OpenAI says it is actively working to strengthen protections so they remain consistent across all exchanges, no matter how long.

Expanding Safety Measures Beyond Threats

OpenAI is also broadening its focus beyond self-harm and violence. The company is exploring how ChatGPT can respond earlier to other risky behaviours-such as dangerous stunts, substance abuse, or extreme sleep deprivation. Instead of waiting for a crisis point, the AI could step in sooner, offering grounding support and suggesting professional help.

For younger users, the company is developing parental controls that give families more oversight of interactions. Another idea under consideration is connecting people in distress with trusted contacts or licensed therapists, so that intervention happens before situations escalate.

Transparency and Privacy

While OpenAI highlights its commitment to user privacy, it also makes clear that privacy has limits when safety is at stake. If a conversation includes plans to harm others, it may be reviewed by moderators and could ultimately trigger real-world intervention, including police involvement.

The blog post serves as a reminder: while ChatGPT is designed to be a safe space for conversation and support, it is not completely private in high-risk situations. Users should understand that threats to others carry consequences both within the platform and, in rare cases, outside of it.